Re: Hand count or open source

[Neal McBurnett <neal@xxxxxxxxxxxxxxxxx>]

It will be much more efficient to hash this out on the wiki site.
It is already discussed there:

http://coloradovoter.net/moin.cgi/StatSampleHandCount

But people are confusing a "statistical sample" with an "audit".  I
want an audit.  ANY discrepancy between the hand count and the machine
count is an ERROR.  In order to be more confident that the software is
not hacked, we just need to look at enough ballots to make sure that
we would be likely to catch fraud/mistakes in action by finding ONE
wrongly-scored ballot.

That's very different from an attempt to estimate how people voted, in
which case the traditional statistics of sample size come into
play.  In this case we don't have to estimate, we have to audit.

Neal McBurnett                 http://bcn.boulder.co.us/~neal/
Signed and/or sealed mail encouraged.  GPG/PGP Keyid: 2C9EBA60

On Fri, Nov 12, 2004 at 08:34:12AM -0700, Nicholas Bernstein wrote:
> You know, there is a whole branch of science dedicated to problems like 
> this; it's called mathematics.
> 
> If you want to be sure that your counts are accurate to within 1% with 
> 95% confidence, you need an unbiased sample size of approximately 10,000 
> for any population size of of 100,000+.
> 
> N
> 
> PS Compare that with the 700 needed for the same accuracy and confidence 
> if people were allowed to check their own votes.
> 
> Some Guy wrote:
> 
> >Statistical sampling is a good idea. Neal proposed 1%. I think that's far
> >too small.